The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a system for processing and interacting with external web content, which presents an attack surface for indirect prompt injection. Malicious content on a website could attempt to influence the agent's logic through parsed data.
Ingestion points: Web content is ingested via TabDriver.findByCss, TabDriver.getByText, and TabDriver.getPageSource() as described in SKILL.md.
Boundary markers: None. The skill does not mention using delimiters or specific instructions to ignore content within the browser DOM.
Capability inventory: The skill allows for navigation (td.get), interaction (element.click, element.sendKeys), and DOM inspection.
Sanitization: The skill explicitly notes in the 'Gotchas' section that sanitization is absent for specific selector methods.
[DYNAMIC_EXECUTION]: The library methods getByText(String) and setComboByDataValue(WebElement, String) perform dynamic generation of XPath and CSS selectors through string interpolation without escaping. As noted in the documentation, this behavior allows for selector injection if the input strings contain control characters (e.g., ' or ]).
[COMMAND_EXECUTION]: The instructions recommend executing the command xattr -dr com.apple.quarantine <chromedriver> on macOS. This command removes the 'quarantine' extended attribute, effectively bypassing macOS Gatekeeper security checks for the downloaded ChromeDriver binary.

com-olexyn-tabdriver